MCP server for interacting with Bitrix24 using natural language and AI dialogues to get data, perform actions, and manage the system.
This server exhibits significant definition quality issues across multiple dimensions. While tool names follow a reasonable verb_noun pattern (get_contact_by_id, create_contact, etc.), descriptions are minimal and non-descriptive, parameter schemas lack critical detail, and output schemas are completely undocumented. The server provides no error handling guidance, no resource composition details, and parameters like 'contact' and 'deal' objects accept generic 'object' types with no field documentation. The codebase shows tool registration via FastMCP decorators, but the actual schema definitions visible in the provided source are skeletal. Many parameter descriptions are trivial translations (e.g., 'Идентификатор контакта' for contact_id) that do not guide LLM usage. No evidence of output schema documentation, pagination support, or chaining field inclusion (e.g., does create_contact return a contact_id that downstream tools accept?). The Russian descriptions and lack of English equivalents also limit accessibility for international use.
Добавление контакта к сделке.
Создание нового контакта.
Создание новой сделки.
Получение контакта по идентификатору.
Получение сделки по идентификатору.
Получение контактов, связанных со сделкой.
Получение списка стадий сделок для указанной категории.
Generic 'object' parameter types with no field documentation. Tools create_contact, update_contact, and create_deal accept a 'contact' or 'deal' object parameter with type 'object' and generic description 'Объект контакта для создания'. No nested field schema, no list of required/optional fields, and no field descriptions. LLMs cannot know what fields to include without reverse-engineering from API docs.
Output schemas completely undocumented. No specification of what fields are returned by any tool. When get_contact_by_id or search_contacts returns results, LLMs have no way to know what fields to extract or how to map contact data to downstream tool parameters. This breaks tool composition and chaining.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 47 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Получение списка контактов с возможностью фильтрации.
Получение списка сделок с возможностью фильтрации.
Удаление контакта из сделки.
Поиск контактов по различным критериям.
Обновление существующего контакта.
Обновление выбранных полей контакта.
Обновление существующей сделки.
Обновление стадии сделки.
Descriptions are minimal, non-English, and lack LLM-optimization guidance. Examples: 'Получение контакта по идентификатору' (raw Russian) and 'Обновление выбранных полей контакта' provide no context on WHEN to use the tool vs similar ones (e.g., update_contact vs update_contact_fields), WHAT the tool returns, or ANY prerequisites. Descriptions should be 50-200 chars, optimized for LLM selection.
Parameter descriptions are trivial translations with no actionable constraints. 'Идентификатор контакта' for contact_id tells the LLM nothing about the valid range, format, or how to obtain it if unknown. No guidance on 'What if the ID doesn't exist?' or 'How do I find a contact ID if I only know the name?'
No pagination or result-limiting guidance. list_contacts and list_deals accept a 'limit' parameter with defaults (50 and 50), but the descriptions do not specify min/max bounds, what happens if limit is exceeded, or whether a next_cursor/total_count is returned. Without this, LLMs cannot reason about batching or result completeness.
No error handling or recovery guidance. Tools are marked READ_ONLY or WRITE but provide no documentation of what errors are possible (e.g., contact not found, permission denied, validation error) or how to recover. Agents have no guidance on 'Should I retry? Ask the user? Call another tool?'
Tool composition broken by missing chaining IDs. If create_contact returns a contact_id, that ID should be documented as part of the response. Similarly, if add_contact_to_deal requires a contact_id, the response from search_contacts must explicitly document that contact_id is returned. No evidence of this in the schema definitions provided.
No distinction between similar tools that could cause LLM confusion. update_contact and update_contact_fields both exist, the description should clearly explain when to use one vs the other. Is update_contact_fields a partial update? Does it merge or replace? No guidance provided.
Security risk: No guidance on permission requirements or scope. Tools are marked WRITE (create_contact, create_deal) or READ_ONLY but provide no documentation of what permissions are required (e.g., 'admin', 'write:contacts', 'write:deals'). This prevents least-privilege configuration and makes audit trails unclear.